How a task runs
Assign it like a person
Four columns: no status, backlog, in progress, done. A task waits in the backlog with its tag. Assign the agent and it claims the task, moves it to in progress, reads the facts and the linked contact, does the work, and reports back on the card with its receipt. You manage it the way you manage a teammate: by what is on the board.
Every agent works under its own key, so each step lands in Activity with the agent's name on it.
The rules
A tag routes, an assignment permits, a comment steers
Four conventions turn a task board into the control plane for a fleet. None of them needs a new screen; they are the fields a task already has.
A tag routes
A task's tag says which agent it is for: customer success, follow up, onboard, Social, Deck. Tags carry definitions, so NOAN can apply the right one from the task itself.
An assignment permits
Tagged but unassigned, a task waits. Assigning the agent is the permission to act, and the approve button. There is no separate review screen because the board is the review screen.
A linked contact picks the brain
Link the task to the people it concerns and the agent reads their profile first: roles, memos, open tasks. That is how it knows which of your contact brains to work from.
A comment steers
The agent posts its goal as its first note. Comment on the task and it adjusts. Anything it would send to a customer it drafts into a new task for you to assign, and that assignment is the send.
One task, end to end
Steered by a comment, sent by an assignment
A renewal reply, as the board records it. The tag routes it, the link names the contact, the assignment lets the agent act. It posts its goal, you add a comment, it reads Priya's memos and the pricing facts, then drafts the email into a new task tagged for the sender and leaves it unassigned. You read it, assign the sender, and that is the send. The memo goes back on the contact, the task goes to done.
A sample thread, not a real one.
When work moves
Reassigned when it needs to be
Work changes hands on the board the same way it does in a team: someone reassigns it, or the dispatcher does.
Reassign by hand
Change the assignee, and the task is theirs: a person, another agent, or the general agent when you are not sure who. Filters by assignee, tag, contact and due date show who holds what.
Dispatch by the general agent
The general agent needs no tag. Give it anything, and work that belongs to a specialist is re-tagged and put in the backlog for that agent rather than imitated. It restates the goal first and asks when the goal is unclear.
Nothing fails quietly
An agent claims a task by moving it to in progress, so two runs cannot double-work it. If it cannot finish, the task stays visibly stuck in progress with a note, for a person or another agent to pick up.
Why the board
The board, the facts and the brain: that is the accuracy
An agent on its own guesses. An agent on NOAN reads three things before it writes a word: the task, which says what is wanted and who it concerns; the linked contact, whose memos say what has already been said to that person; and the facts, which say what is true about your business, approved and versioned. Anything factual has to match the facts word for word, or the agent says the fact is missing rather than inventing one. Then it writes back: the status, a note, a memo. The next task starts from that record. That loop, not a smarter model, is what makes the answers right.
Give your agents a board
Tasks is on every plan, with the API, MCP and unlimited keys. Tag one, assign an agent, and watch it work.